I am trying to find out how inflammatory markers change in response to a mechanical device change. I have 5 pre readings and 5 post readings. I am expecting a rise in inflammatory markers in the pre readings followed by a fall after the device change. I also need to find out how the change differs in two subgroups. The aim is to find out if the change is significant to consider these markers as predictors of the device failing and hence needed to be changed. I am a non-statistician and have access to SPSS. I have some understanding of the software. I would like to know what test would be appropriate to analyse this. I need to keep things simple as people reading the results are non-statisticans as well. I would really appreciate any advice.
You can use within-subject correlation method. See Bland and Altman's paper for detail here https://www.bmj.com/content/310/6977/446
Your design sounds like a paired before-after control-treatment (PBACI). There are a few ways to conduct the analysis (e.g., randomized intervention analysis, mixed-modeling, t-test).
I think Conner et al., 2016 do a good job of offering an overview for BACI design and analyses, provide good references, and present an useful alternative approach to analysis and interpretation. The paper is written to a different field, but I think it's easily exchangeable.
https://doi.org/10.1007/s10661-016-5526-6
Alternate access: https://rdcu.be/bFNLX
As the above answers suggest data like this have been modeled different ways in the past. My first choice would be longitudinal models. The best description is given here:
https://www4.stat.ncsu.edu/~davidian/st732/notes.html
SPSS is a poor resource for these models. See the discussion in the notes for better advice. Best, David Booth
Thank you for your help.
I need further advice please. So I got help from someone to do a linear mixed model to look at significance of change in the markers before and after the device change. However, a post hoc test was not done at the time of the testing to find exactly where the differences are. Will it be appropriate to do a a paired t-test between time points? Or will a graphical display with a line graph suffice?
A linear mixed model is I believe equivalent to what a biostatistician calls a longitudinal model. It depends on where you submit. I suspect you should use a hypothesis test. Since you have only two time points I believe that your t test would be equivalent to the longitudinal model's formulation. I however am not an expert. I suggest you consult someone who is. Best regards, David Booth
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS